Our TB association study extends across two genetically highly diverse populations. It combines GWAS in Indonesian population and follow-up genotyping of the best associated SNPs in a large independent cohort from Russia. Our data provide evidence of possible novel associations of genetic variants with pulmonary TB susceptibility. Among them (Table 2), one of our lowest P values is observed for rs2273061 (P 0.0004, O.R 1.16, 95%C.I. 1.07-1.26), which is in the transcript of JAG1. This protein is a ligand for the Notch receptor that plays a central part of the Notch signaling cascade 27 . Mouse macrophages infected with M. bovis BCG have been shown to up-regulate NOTCH1 signaling leading to SOCS3 expression via NOTCH1 mediated recruitment of NFκB and CSL to the SOCS3 promoter. As SOCS3 is a critical regulator of cytokine signaling, induction of this gene by mycobacteria could suggest a strategy to render infected macrophages unresponsive to interferon-gamma (IFN-γ), which is a central Th1 cytokine 28 . This modulation of host cell signaling response may be critical for a generalized suppression of inflammatory responses, and the persistence of mycobacteria within the host.

Notch signaling also plays a pivotal role in T cell lineage commitment, and another associated SNP, rs10515787 (P 0.0013, O.R. 0.79, 95%C.I. 0.68-0.91) is in the EBF1 gene, which is a central B cell lineage specification factor. In order for EBF1 to perform its role, it must partner with PAX5 through a feedback regulation to amplify B cell specific gene expression and solidify the commitment to the B cell pathway 29 . PAX5 is the guardian of B cell identity and functions by down regulating genes that are against B cell lineage, such as the M-CSF and NOTCH1, which are required for myeloid development and T cell lineage specification respectively 30 31 . This counteractive response of repressing NOTCH1 signaling that is not in favor of T cell promotion, might suggests an impact on the control of the intracellular infection of M. tuberculosis.

Two SNPs rs10497744 (P 0.0014, OR 0.87, 95%C.I. 0.80-0.95) and rs1020941 (P 0.0022, OR 0.88, 95%C.I. 0.81-0.95) in LD (r ² = 0.99, D' = 1.00) that are part of the associated list are near the TMEFF2 gene. This gene encodes a transmembrane protein with EGF (epidermal growth factor)-like and two follistatin-like domains 2, which is known to contribute to cell proliferation. Shedding of TMEFF2 from the ectodomain is a functionally important step to release the protein in its active form for inducing cellular proliferation. This functionally limiting step is highly mediated through an ADAM17 dependent autocrine fashion 32 . Incidentally, ADAM17 also has a prominent role in activating the cell-fate specification Notch signaling pathway, by controlling the shedding of Notch receptor and its ligand JAG1 33 , which is also our first target gene, mentioned above. An active ADAM17 regulates EGF receptor expression through activating NOTCH1 that was demonstrated to affect proliferation and survival of lung cancer cells, and tumorigenicity of non-small cell lung cancer 34 . However, on the other hand, inactivating NOTCH1 or ADAM17 resulted in substantial cell death, while EGFR inhibition predominantly induced cell arrest in lung cancer 34 . Studies have also shown ADAM17 actively mediates the shedding of pro-inflammatory factors in lung inflammation, and regulate immune cell recruitment and cytokines secretion that affects the physiology of this organ 35 . In pulmonary TB, the lung is the primary site of infection by M. tuberculosis where, responding to invasion, our body reacts by recruiting immune cells and pro-inflammatory cytokines to attack and control further dissemination of a pathogen by forming granuloma, which may also manifest in tissue damage.

Another example of biologically relevant candidate in our data is rs188872 (P 0.0023, O.R. 0.87, 95%C.I. 0.80-0.95), which is near the CCL17 cytokine gene. Lung granulomas in mice were reported to have enhanced CCL17 transcript levels after being stimulated with M. bovis antigen 36 . As a survival mechanism, pathogens such as M. tuberculosis are known to preferentially shift host cell response towards Th2 by instigating the production of Th2 cytokines. When in excess, it would consequently lead to immuno-suppression that might antagonize the Th1 mediated microbicidal actions. In natural infection M. tuberculosis may likely gain from this favorable condition to survive in infected patients.

Within the Indonesian study, the lowest P value is found in rs10497225 (P 1.52 × 10^-5, O.R. 2.36, 95%C.I. 1.58-3.52), which is in the SLC4A10 gene, see Additional file 2, Supplementary Table S1. This solute carrier family 4, sodium bicarbonate transporter, member 10 (SLC4A10) gene is in a similar class of function as the ion transporter; SLC11A1 (alias NRAMP1), a well studied TB gene involved in iron metabolism and host resistance to pathogenic mycobacteria. Genetic variants of this gene have been associated with susceptibility to TB and leprosy 37 38 . However, we could not analyze rs10497225 in the Russian cohort because this SNP is rare (MAF 0.0007) in this population, and was excluded after failing MAF filter. In view of this, we believe some of the association signals could be affected by possible genetic differences between the host populations. As these SNPs are merely markers tagging the actual causal variants based on linkage disequilibrium (LD), differences in LD patterns and allele frequencies between differing ethnicities could affect the efficiency of transferring tags across populations and the power in detecting associations. This is notwithstanding the fact that the 100 K SNP GeneChip marker set used in Stage1 is a rather sparse collection of SNPs. The SNPs in this microarray capture (r ² ≥ 0.8) common variants in the Asian (JPT+CHB) and European (CEU) genomes at only 30% coverage 39 , that are also undersampled in the coding regions, reducing the level of proxy to genes 40 . Hence, it is likely that certain regions in the genome are less adequately tagged with SNPs, which could thereby have resulted in reduced power for detecting associations.

Although none of the observed association signals achieved stringent levels of genome wide significance, likely due to the limited sample size of the Indonesian GWAS cohort, the major findings from the study of both Indonesians and Russians does suggest associations at several loci, many of which are located in, or close to immune related genes that have congruous functions toward Th1 axis of the pro-inflammatory IFN-γ activity. IFN-γ is an essential cytokine for the effective control of M. tuberculosis in the host, due to its central role in modulating and bridging both the innate and adaptive immunity, impairments in this axis of cytokine activity could render adverse consequences. A previous study conducted on a subset of samples from the same Indonesian cohort, had peripheral blood cells taken from active TB patients, patients undergoing treatment, and healthy controls, and traced for Th1 cytokines production in response to M. tuberculosis and mitogen stimulations 12 . The integrity of major pathways involved in Th1 immunity were analyzed, among them IFN-γ level was found to be significantly correlated with TB disease activity and response to curative treatment, that was specific to M. tuberculosis stimulation 12 . This change in cytokine activity according to the disease course of pulmonary TB is unlikely due to major defects in IFN-γ itself, since mutations in this molecule and its receptors are known to implicate rare severe infections to otherwise poorly pathogenic mycobacteria 41 42 . Rather, in pulmonary TB, a complex disease with adult onset, it is more likely due to the accumulation of individual subtle effects from variations in genes, such as those suggested from this study that are working together in similar pathways, which might sway the immune responses of the group of susceptible individuals toward active disease.

Tuberculosis is a complex disease resulting from multiple contributing factors, and the mechanism that triggers active disease is unlikely to be simplistic. Aiming to expand TB disease knowledge, this study took a comprehensive search across the genome, and suggests multiple targets working in novel pathways involved in the host containment of infection with TB, further providing insights on the mechanism of this disease, that could previously be neglected in hypothesis driven approach.

TB: tuberculosis; GWAS: genome wide association scan; SNP: single nucleotide polymorphism; HIV: Human immunodeficiency virus; PC: principal component; MAF: minor allele frequency; HWE: Hardy Weinberg equilibrium; QC: quality control; WHO: World Health Organization; OPA: oligo pool assay; IBS- identity by state; LD- linkage disequilibrium; OR- Odds ratio; QQ plot- quantile-quantile plot; λGC- lambda genomic control inflation factor; CMH: Cochran-Mantel-Haenszel; JPT: HapMap Japanese from Tokyo; CHB: HapMap Han Chinese from Beijing; CEU: HapMap Caucasian from North America.

The authors declare that they have no competing interests.

All authors contributed in various phases to the writing, and had read and approved the final manuscript. TO and MS were the principal investigators of the study, and supervised it throughout together with RC and MH. BA, ES, RN all played crucial roles in patient and control selection and sampling. EP performed the genotyping, statistical analysis, and the drafting of this manuscript. EV contributed to many discussions and helped writing the manuscript. IA contributed in processing biological samples and managing the database. SM was instrumental in co-designing the project. YB, VN, FD co-ordinate and implemented patient and control selection and sampling sample in Russia. SN participated in sample collection in Russia and association analysis of the Russian data.